Antropic blocks users who insult AI… “Conversation will end if Claude continues to abuse”

(Seoul = Yonhap News) Reporter Go Il-hwan = In the future, use of Claude, the generative AI from artificial intelligence (AI) company Antropic, may be blocked if you continuously swear or use unnecessary insults.

According to the BBC on the 9th (local time), Antropic revised its usage policy the previous day to allow users to end conversations if they engage in "insulting or cruel behavior" toward AI.

Antropic explained that it would only apply the new policy to repetitive and extreme cases.

This means that it will not be applied to general user complaints or objections, model testing and research, etc.

However, Antropic did not specify which actions would be judged as insulting or cruel toward AI.

Currently, Antropic's list of prohibited activities includes harassing others, encouraging self-harm, and producing non-consensual sexual images.

Regarding Antropic's revision of its usage policy, controversy arose on social media over the extent to which AI should be treated like humans.

Some users evaluated it as an act to encourage proper manners, but some also criticized AI's excessive anthropomorphism.

“This level of anthropomorphism of AI is harmful,” said Dr. Barry Scannell of Irish law firm William Fry. “It can lead people to mistake AI for something other than what it really is.”

Criticism has also been raised that the meaning of actual abuse targeting humans or animals may be diluted.

At the same time, debate continues over whether polite expressions such as ‘please’ or ‘thank you’ are necessary when talking to generative AI.

Some experts point out that this type of expression unnecessarily consumes the tokens used to process AI responses.

On the other hand, there is a counter-argument that polite expressions can lead to better answers.

Last year, when a user asked about the power cost for ChatGPT to respond to users' "please" and "thank you," OpenAI CEO Sam Altman responded, "It costs tens of millions of dollars, but it's money well spent."

Meanwhile, in the process of revising its usage policy, Antropic stipulated that the use of Claude to deceive voters and interfere with elections is also prohibited.